hanetzer, Kamil Debski various presentation of when he was designing it are good start, even though things have evolved a bit since then
ndufresne: hello andy :)
hanetzer, and then exsiting driver like venus, coda, then there is more complex driver like the original mfc
there work happening right now on Amlogic, Ely got some frames out now using ffmpeg and GStreamer
noice
what are you interested in ?
ndufresne: some grain media SoC's with a lot of faraday ip on it.
(oh, I didn't mention that there is work on Allwinner and Rockchip to support stateless codecs)
they have jpeg/mjpeg/h264 stuff
stateful or stateless ?
what do either of those mean?
I know stuff, but not a lot of exact terminology :P
state full codec are simplier, you just through frames to it, and it manages everything, you get some frame out at some point
state less requires userspace to parse the bitstream and fill giant structures, then you need to allocate some space for the "state" and pass this over and over
huh. I honestly have no idea.
usually, statefull codecs comes with a firmware, often a control software running on a co-processor
ndufresne: ah, like vc4 or whatever the raspi is
so you have a reference implementation or some programmer manual of any kind ?
yeah. vendor code and register definitions
Well, yeah, except the co-processor on RPi is bigger then usual
its involved in booting the main proc, right?
that I don't know, it's a baseband, it runs it's own OS and manages everything multimedia, GL, rendering to screen, capturing from camera, codecs, audio etc.
VC4 driver breaks the paradigm, the display and GL control is put back on the main core
yeah, something funky like that.
and there is an IPC and a bus to communicate with the baseband, called mmap
mmal
but all this isn't relevant for your work I guess ;-P
nah. honestly this is a bit far off, still need to get a more recent kernel booting on it :P
thankfully the vendor kernel isn't a 2.6.x line. 3.3, which is bad, but not that bad
hanetzer, maybe you should come by #linux-amlogic, and chat with Ely, he's doing the same as you, there is zero-doc for this, just a full of global variable spagetti driver, Ely has been picking small parts, into a proper v4l2 m2m driver, loading firmwares at first, then hardcoding stuff to get 1 frame decoded, ...
all this by picking code from the downstream driver, 3.10 and 3.14, which is not using any of the framework
hanetzer, implementing v4l2 m2m is interesting, because you get generic userspace for free, notbably ffmpeg and gstreamer
(and chromium/chrome)
nice. those wouldn't be useful for chro*, its an ip camera soc
suppose it could be used for other stuff, but that's what its marketed for :P
however, thanks for the pointers and info. I'll pseudo-afk for now, time to write some spi flash drivers :)
or rather, translate vendor shitware spi flash controller drivers into something u-boot can accept as mainline
a lot of ip cameras runs gstreamer, ever heard about Axis ?
I'm aware of the name, yes. but I wasn't speaking of gstreamer. chro*->chrom{e,ium}
ah chrome, right, would be weird to have chrome on an ip camera ;-P
ffmpeg or gst, more likely
I've seen some strange things in this field, to be fair.
how known when someone will driver an LCD on a camera and implement the UI using a webapp
have fun with u-boot, sounds like it will be really useful !
well, I do sort of intend on doing something like that for DVR/NVR's if I can find one to work on that isn't based on farking hisilicon chips. blobs as far as the eye can see
29 blob kernel modules
per soc
and they apparently have a crapload of them
<<^^>> note the above stuff, hanetzer
don't mind that, just a bookmark in my irclogs :)
Have some helpers made it into vb2 regarding cpu cache management for dma buffers ? Let's say we want cpu read cache enabled on them, and we flush it right before userspace dequeues a CAPTURE buffer, is that something that can be easily achieved now ?